Thesis
Trust repair strategies in conversational search
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2025
- Thesis identifier
- T17568
- Person Identifier (Local)
- 202081697
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- The rapid integration of conversational artificial intelligence into financial services promises to transform customer engagement by delivering on-demand support and automating routine tasks. However, user trust remains fragile, especially when chatbots err. This thesis investigates trust dynamics in financial chatbots using three controlled experimental studies involving a Microsoft Azure based chatbot prototype. We examine how different types and frequencies of errors undermine trust, how targeted repair strategies can restore it, and how individual personality differences shape both trust breakdown and repair effectiveness. We also explore the stabilising role of chatbot benevolence, expressed through personalisation and empathy. The rapid integration of conversational artificial intelligence into financial services promises to transform customer engagement by delivering on demand support and automating routine tasks. However, user trust remains fragile, especially when chatbots err. This thesis investigates trust dynamics in financial chatbots using three controlled experimental studies involving a Microsoft Azure based chatbot prototype. We examine how different types and frequencies of errors undermine trust, how targeted repair strategies can restore it, and how individual personality differences shape both trust breakdown and repair effectiveness. We also explore the stabilising role of chatbot benevolence, expressed through personalisation and empathy. Drawing on these experiments, we first quantify trust degradation across error conditions, factual inaccuracies, misinterpretations, and delayed responses and identify tolerance thresholds beyond which trust collapse becomes unlikely. Next, we isolate the impact of benevolent behaviours on trust formation and maintenance, demonstrating that empathy and personalised content significantly buffer against minor failures. Finally, we assess how the Big Five personality traits (Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism) moderate responses to affective (apology), functional (compensation), and informational (explanation) trust repair strategies. A personality-aware random forest model predicts the most effective repair tactic with 73.4% accuracy. We synthesise these findings into an integrated framework comprising four perspectives the Trust Dynamics Cycle, Ecological System, Interaction Attribution, and Dual Process model, and propose novel theoretical contributions: Trust Resilience Theory, Dual Threshold Model of Collapse, Personality-Matched Repair Strategy Theory, and Benevolence Accuracy Balance Theory. The results yield concrete design principles for developing financial chatbots that adapt repair strategies to user dispositions, calibrate benevolence signals to error severity, and maintain robust trust even when conversational errors occur.
- Advisor / supervisor
- Ruthven, Ian, 1968-
- Moshfegi, Yashar
- Resource Type
- DOI
Relations
Contenu
| La vignette | Titre | Date de téléchargement | Visibilité | actes |
|---|---|---|---|---|
|
|
PDF of thesis T17568 | 2026-01-21 | Public | Télécharger |